Paige, a leader in AI-driven pathology solutions for cancer, announced the open-source release of its Foundation Models, Open Paige Foundation Model (OpenPFM), featuring Virchow and PRISM. These pioneering models, now available on Hugging Face, are designed to advance cancer research and clinical development. Virchow is the first million-slide foundation model for cancer, while PRISM, building on Virchow, is a vision and language model for comprehensive slide-level cancer insights. An ML SDK is also provided to support their use in the research community.
Developed with Microsoft Research, Virchow and PRISM aim to unlock deeper insights within pathology slides. Virchow's training data includes 1.5 million digitized slides from Memorial Sloan Kettering Cancer Center, one of the top cancer hospitals in the U.S. Published in Nature Medicine, Virchow sets a new standard in computational pathology, aiming to revolutionize cancer diagnosis and treatment.
PRISM enhances Virchow’s image representations, creating whole slide image signatures and generating written diagnostic summaries. Trained with an additional 587,000 slide images and 195,000 clinical reports, PRISM excels in tasks like cancer detection and subtyping without extra training. It improves data efficiency, reducing the number of training samples needed, and shows promise in biomarker prediction across various tissues.
Siqi Liu, Director of AI Science at Paige, emphasized the company’s commitment to fostering innovation and accessibility in AI research. Razik Yousfi, SVP of Technology at Paige, highlighted the goal of driving innovation and enhancing cancer diagnostics and drug development to transform patient care. Virchow is available under the Apache 2.0 License, while PRISM and the ML SDK are under the CC-BY-NC-ND 4.0 License. Advanced versions and commercial licensing options are available for clinical trials and R&D. Paige also offers collaboration opportunities to further scientific discoveries.